Dynamics - Lecture Notes

Chris White ([email protected])

Abstract
These notes are a write-up of the rst half of the lecture course Dynamics, Electrodynamics
and Relativity, as given at Glasgow University in September / October 2010. Given that there
were two separate lecturers for each part of the course, these notes cover only classical (non-
relativistic) dynamics.
1 Introduction
Classical dynamics is by now an old subject, i.e. at least several hundred years old if one takes
Newtons Laws as a starting point. As such, classical physics does not always seem as trendy
as more modern developments (special / general relativity, quantum mechanics etc.). Neverthe-
less, classical non-relativistic dynamics is a highly complex and beautiful subject, and is still
providing new areas of active research (e.g. chaos theory is a fairly young oshoot). Further-
more, the ideas and formalisms of classical physics underly all other modern theories of physics.
There is a strong sense in which the language of modern physics is deeply rooted in classical
physics, particularly in the Lagrangian formalism, which we will learn in this course. Thus, an
understanding of classical behaviour is a necessary prerequisite for understanding systems which
are quantum, relativistic, or both! What I hope we will also see is that classical mechanics is
interesting in its own right. After all, we are all surrounded by classical phenomena in everyday
life!
When we think of classical mechanics, this almost immediately conjures up Newtons Laws
of motion, which one may state as follows:
1. Objects at constant velocity remain so if no forces act.
2. The force f experienced by an object governs its acceleration a by the relationship
f = ma m
d
2
x
dt
2
, (1)
where m is the inertial mass of the object, and x its position.
3. If object A exerts a force f on object B, then object B exerts a force f on object A. This
is sometimes summarised as every action has an equal and opposite reaction.
The rst law seems obvious nowadays, although perhaps only because we have it drummed into
us from an early age. Nevertheless, the idea that objects can continue moving without any
applied force would certainly have seemed strange to pre-Newtonian physicists (including the
ancient Greeks), perhaps because in everyday life motion is usually damped by air resistance or
friction. Thus, the rst law is more of a conceptual breakthrough than it might rst appear.
The second law is a clear statement of how a particles trajectory (x(t)) is related to the
forces which act on the particle. It gives a dierential equation for this trajectory, and tells us
that if we know the initial conditions of a particle at some initial time t
0
(e.g. its position and
velocity), we can calculate the trajectory at all later times. If there are many particles, each
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
acted upon by forces, we have one such equation for each particle. These are the equations of
motion of the system.
The third law is perhaps the most confusing and misunderstood, but often helps us to sim-
plify calculations in many-particle systems. We will see this in what follows.
Newtons Laws are relatively easy to apply to simple mechanical systems e.g. one or two
particles, with a force acting. It can also be useful in many-particle systems, where the equations
of motion can be solved numerically if necessary. However, for very complicated systems (e.g.
particles subject to nontrivial constraints) Newtons Laws can be dicult to apply. For some
systems, e.g. elds rather than particles, it is even more confusing to have to think in terms of
forces. Thus, it is useful to have powerful, general methods for describing classical dynamics.
The aim of this course is twofold. Firstly, we review some aspects of Newtonian mechanics
which may be unfamiliar (e.g. rotating / accelerating reference frames, many particle systems).
Secondly, we develop an alternative formulation of classical dynamics, which avoids us having to
think about forces at all, and which is much easier to apply to complex mechanical systems. This
is the Lagrangian formalism, and we will see that this approach not only makes problem solving
easier, but also reveals hitherto hidden conceptual ideas of great power (e.g. the relationship
between conserved quantities and symmetries).
2 Rotating Reference Frames
2.1 Inertial frames
A frame S is a system of coordinates used to describe the motion of a system. Each object has
coordinates (x, y, z) in such a frame (assuming 3 spatial dimensions), where each of these de-
pends on time t if the object is moving relative to the frame S. Let (x

, y

, z

) be the coordinates
of a given object in S

. These will be related by some transformation to the coordinates in S,

which may be quite complicated in general, depending on how the frames are moving relative
to one another.
An inertial frame S is a frame in which Newtons rst law is obeyed. Equivalently, an inertial
frame is a frame moving at constant velocity relative to other inertial frames i.e. objects at rest
in such a frame will remain so by Newtons rst law. The latter denition is somewhat circular
- we can give yet another denition later.
Now consider an inertial frame S

which is moving with speed v relative to another frame S,

where coordinates in each frame are labelled as above. Without loss of generality, we may take
this motion as being in the x direction, such that the frames coincide precisely at time t = 0
(gure 1). It is not too dicult to show that the coordinates in S

are related to those in S by

the following transformation:
x

= x vt
y

= y
z

= z
t

= t (2)
This is an example of a Galilean transformation, the name given to transformations between
inertial frames (more generally, one can also have rotations and translations relating S and S

,
in addition to a constant velocity vector v in an arbitrary direction). Note that we have also
included the time transformation for completeness. Here the times coincide as a result of the
2
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 1: Two inertial frames S and S

, where S

moves with constant speed v along the +x

direction relative to S. The two sets of coordinate axes coincide at time t = 0, and the z and z

axes are not shown.

coordinate axes being coincident at t = t

= 0. In general, one could have an additive constant

relating the times in the two frames. However, this is the most general transformation for the
time coordinate, and this constant merely reects the fact that we can choose what we call zero
time in any given frame (i.e. one is free to reset clocks at will). From now on, we set this
constant to zero, so that the time coordinate t is the same in all frames.
The above discussion highlights an important point about Newtonian mechanics: time is
universal, or absolute. By this we mean that the passage of time is the same for all observers.
Dierent observers may disagree on the exact reading on their clocks up to an arbitrary shift
(an example of this is time zones in dierent parts of the world), but the rate at which a clock
ticks is the same in all frames. This will not be true later in the course, when you go beyond
Newtonian mechanics.
Now consider a particle of mass m moving through space. If its position as measured in S is
x(t), then the velocity and acceleration in this frame are x(t) and x(t), where the dot represents
the derivative with respect to time. The corresponding quantities in S

are x

(t), x

(t) and x

(t),
and using the above transformation one nds
dx
dt
=
_
_
x
y
z
_
_
=
_
_
x

+ v
y

_
_
=
_
_
x

_
_
+
_
_
v
0
0
_
_
. (3)
We can interpret the nal result as follows:
(Velocity in S)=(Velocity in S

)+(Velocity of S

relative to S)
This is the velocity addition rule of Newtonian mechanics, and it tells us how velocities trans-
form between inertial frames. We also recognise this behaviour from everyday life. Things are
more complicated when transforming between non-inertial frames, as we will see in the following
3
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
section.
Similarly as for the velocities, we may also nd how accelerations transform. From the above
transformation, one nds that the accelerations in S and S

are related by
d
2
x
dt
2
=
_
_
x

_
_
=
d
2
x

dt
2
(4)
That is, the acceleration of the object is the same in both inertial frames. Note that this relied
crucially on the fact that the speed v (of S

relative to S) was constant. That is, both frames

have to be inertial. Assuming that the mass and force are also the same in both frames, this in
turn implies that Newtons second law is obeyed in both inertial frames. The general expression
of this idea is
The laws of physics are the same in all inertial frames.
This is the principle of Galilean relativity. It tells us that only relative motion has any
meaning, and that there is a special class of frames in which this is most clearly observed -
those picked out by Newtons rst law. We can use the above principle to give us yet another
denition of an inertial frame i.e. a frame in which Newtons second law is obeyed.
The above discussion has summarised Newtonian mechanics, from a slightly formal view-
point - that of Galilean relativity. The relativity principle was, historically, a huge conceptual
breakthrough in understanding nature. We now know that Galilean transformations are only
approximate, and break down when objects move suciently fast (i.e. at speeds close to the
speed of light). Then, special relativistic mechanics takes over, with Lorentz transformations
replacing Galilean transformations. Nevertheless, the relativity principle (and the special nature
of inertial frames) survive in the special theory of relativity. The latter diers from Newtonian
mechanics in that there exists a speed (the speed of light) that is the same for all inertial ob-
servers - a fact clearly in contradiction with the velocity addition law of Galilean relativity that
we derived above. You will see how this works in the second half of this lecture course. For
now, let us return to Newtonian mechanics.
So far we have only discussed inertial frames, and we have seen that the Newtons laws are
the same in all such frames. This does not mean that we cannot do physics in accelerating
(non-inertial) frames. Rather, there will be modications to the laws of motion in non-inertial
frames whose origin can be traced to the fact that such frames are accelerating. A classic
example of this is motion near the Earths surface, which (although it looks locally very like an
inertial frame) is aected slightly by the rotation of the Earth. We have seen that Newtons
second law f = ma is guaranteed to hold in all inertial frames. In a general non-inertial frame,
it perhaps follows that there will be modications to this law. That is, extra terms will appear
in Newtons equation, which we will be able to interpret as ctitious forces. In order to make
all of this more concrete, it pays to study a particular example - that of rotating frames. This
is the subject of the next section.
2.2 Rotating frames
A rotating reference frame (with a constant angular velocity about some axis) is one of the
simplest examples we can think of of a non-inertial frame. Nevertheless, it is already sucient
to illustrate some of the more confusing aspects of classical mechanics in non-inertial frames...!
4
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 2: A frame S
1
(coordinates (x
1
, y
1
, z
1
)) which is rotating relative to a frame S
0
about the z
0
axis (coordinates (x
0
, y
0
, z
0
)) with angular velocity = (0, 0, ), such that the axes are coincident
at time t = 0. The angle between the frames is = t, and the z
1
, z
0
axes are not shown but
coincide at all times.
It is also a particularly relevant example, given that coordinates on the Earth constitute a
rotating frame.
To be specic, we consider a frame S
1
rotating with angular velocity about the z-axis
relative to an inertial frame S
0
, such that the frames coincide at time t = 0. That is, the vector
angular velocity of the frame S
1
is = (0, 0, ), so that the x
1
and y
1
axes (of S
1
) rotate
anticlockwise about the z axis (of S
0
), if one looks down the z axis. This setup is shown in
gure 2. Our eventual aim will be to see how velocities, accelerations etc. look in the two
frames. For now, let us consider some vector whose components in S
0
and S
1
are given by A
0
and A
1
respectively. This may represent e.g. a position of an object, and thus in general will
depend upon time t in both frames. At any given moment, however, the components of the
vector in S
0
are related to the components in S
1
by the transformation:
A
0
= R()A
1
, (5)
where is the angle between the frames as shown in gure 2, and R is a rotation matrix. For
example, for rotations about the z axis, this has the explicit form
R =
_
_
cos sin 0
sin cos 0
0 0 1
_
_
, (6)
as can be veried by acting on unit vectors along the x
1
, y
1
and z
1
axes. At time t one has
= t, so that one may rewrite eq. (5) as
A
0
(t) = R(t)A
1
(t), (7)
where we have also made explicit the fact that both sets of components depend on time in
general. It follows from this equation that the rate of change of the vector as seen in S
0
is not
the same as that seen in S
1
. This may seem confusing at rst, but it is in fact perfectly natural
- the rate of change of the vector as seen in S
0
has two contributions, due to both the change in
S
1
, and the rotation of the coordinate axes of S
1
. The easiest way to think about this is perhaps
to imagine a vector which is xed in S
1
, so that

A
1
= 0. However, by eq. (7), the vector in S
0
is seen to rotate about the z axis with angular velocity , so that

A
0
is denitely not zero!
5
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 3: Change in the axes of S
1
in the timestep t t + dt.
In more detail, let us consider the timestep t t + dt, during which A
0
A
0
+ dA
0
and
A
1
A
1
+ dA
1
, where
dA
i
=
dA
i
dt
dt. (8)
From eq. (7) one has, using the product rule,
dA
0
= [dR(t)]A
1
+R(t)dA
1
, (9)
which relates the change of the vector as seen in S
0
with that seen in S
1
. There are indeed two
terms on the right-hand side. The second represents the change due to the fact that A
1
changes
in S
1
, appropriately transformed to S
0
using the rotation matrix (this term vanishes for vectors
which are xed in S
1
). The rst term represents the change in S
0
that results due to the fact
that the axes of S
1
are rotating relative to S
0
. In fact one may show that
[dR(t)]A
1
= A
0
dt
= [R(t)A
1
]dt, (10)
where is the vector cross product, and we have used eq. (7) in the second line. One can show
this purely using the appropriate form of the rotation matrix (eq. (6)) - see problem 1.1. Here
we show this result using hopefully more intuitive arguments.
The term [dR(t)]A
1
represents the change in A
0
that is purely due to the rotation of the
axes of S
1
. In the timestep we are considering, the axes move as shown in gure 3. This induces
a change in A
0
, which (given that dt is innitesimal) is perpendicular to A
0
, as shown in gure 4.
From the geometry of the gure, one nds
|dRA
1
| = dt|A
0
(t)| = | A
0
|dt, (11)
where the second equality follows due to the fact that the change in A
0
is perpendicular to A
0
for an innitesimal time interval dt. Using the right-hand rule for cross-products, one can also
see that the vector A
0
points in the same direction as dRA
1
(recall that the vector angular
velocity points in the positive z direction). Thus, one may remove the magnitude signs (| . . . |)
in eq. (11), which indeed gives eq. (10).
The more formal argument (which is explored on the problem sheet) for this result is that
operation A
0
is equivalent to an innitesimal rotation about the rotation axis (in this case
the z axis), a result one can obtain directly from the explicit form of the rotation matrix. Note
that this is also what we have really done above - we used the geometry of gure 4 to work out
6
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 4: Change induced in A
0
by the rotation of the axes of S
1
. As dt 0 the change becomes
perpendicular to A
0
at time t.
the eect of an innitesimal rotation of the axes. Given that this rotation is about the same
axis as the nite rotation R(t), this allows to rewrite the second line of eq. (10) as
[RA
1
] = R[ A
1
]. (12)
In words: the innitesimal rotation of a nite rotation of A
1
is the same as these rotations
applied the other way around (provided these rotations are about the same axis, which is true
in this case). Thus, one may write
dR(t)A
1
= R(t)[ A
1
]dt, (13)
and combining this with the other term in eq. (9) one gets
dA
0
= R(t)[dA
1
+ A
1
dt]. (14)
Finally, dividing by dt, we nd

A
0
= R(t)[

A
1
+ A
1
]
This important equation relates time derivatives of vectors between the rotating frame and the
inertial frame. The most important thing to note is that this is not the usual transformation
law of eq. (7)!!! Put more simply: time derivatives of vectors do not transform in the same way
that vectors do, so one must be rather careful. The extra term in eq. (14) has important
physical consequences, as we discuss in the next section.
2.3 Coriolis and Centrifugal forces
In the previous section, we derived how time derivatives of vectors transform between rotating
and inertial frames. We saw that this transformation law is not the same as that for the vectors
themselves. Rather, there was an extra term which represented the change in a vector as seen
in the inertial frame, due to the rotation of the axes of the non-inertial frame. The aim of this
section is to use this result to show how Newtons second law is modied in the rotating frame,
thus providing a direct example of the fact that the principle of Galilean relativity does not hold
for non-inertial frames.
We again consider the frame S
1
rotating with respect to an inertial frame S
0
, as discussed
in the previous section and illustrated in gure 2. Let a given particle have positions x
0
and x
1
in S
0
and S
1
respectively. From eq. (7) these are related by
x
0
= R(t)x
1
. (15)
7
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
The velocities transform dierently. By substituting A
0
= x
0
and A
1
= x
1
in the transformation
law derived previously, we nd
x
0
= R(t)[ x
1
+ x
1
], (16)
or
v
0
= R(t)[v
1
+ x
1
]. (17)
This equation has the same form as eq. (7), so that by setting A
0
= v
0
and A
1
= v
1
+ x
1
in the same transformation law as before, one nds
v
0
= R(t)[ v
1
+ x
1
+ (v
1
+ x
1
)]
= R(t)[ v
1
+ 2 v
1
+ ( x
1
)]. (18)
Recognising v
i
= a
i
as the acceleration observed in frame S
i
, one nds
a
1
= R(t)a
0
2 v
1
( x
1
)
Thus, the acceleration observed in the rotating frame S
1
is not the same as that observed in
the inertial frame S
0
. There is a term representing the appropriate rotation of a
0
, which would
also be present in the case of rotated inertial frames (i.e. with constant angle). However, there
are two extra terms whose origin is related to the rotation of S
1
(we can see this directly, as these
terms vanish if = 0). This contrasts with the case of transformations between inertial frames,
in which the acceleration was indeed the same for all observers. This ensured that Newtons
laws were the same in all inertial frames, a fact which gives us a clue about how to interpret the
extra terms appearing in the acceleration transformation equation for rotating frames.
We can still use Newtons second law in both frames, by dening the force on the particle to
be that which satises
f
i
= ma
i
(19)
in frame S
i
. Crucially, the force now depends upon the frame (because the acceleration does).
In the inertial frame S
0
, the force f
i
just consists of the genuine force which are acting on
the particle. In the frame S
1
, we have the appropriate rotation of this force, together with
two ctitious forces, which we can read o after combining eq. (19) with the acceleration
transformation law. These are:
f
Cor.
= 2m v
1
; (20)
f
Cent.
= m ( x
1
). (21)
The rst of these is the so-called Coriolis force. It is proportional to the velocity of the particle
in the rotating frame (as well as the mass and the angular velocity of the rotation). Further-
more, the force is perpendicular to the direction of motion of the particle, and is zero if v .
The second force is the centrifugal force, possibly one of the most misunderstood concepts
in all of physics! The force points perpendicularly outwards from the rotation axis, and is zero
if the particle has no displacement transverse to this axis. Furthermore, the force goes like the
square of the angular velocity.
Although we have used the name ctitious forces to denote that these forces are artifacts
of the use of a non-inertial frame, it must be stressed that in the rotating frame, the Coriolis and
centrifugal forces are for all intents and purposes real forces, with genuine physical eects. It is
perhaps useful to consider a simple example of a system viewed in both frames, to show that the
pictures are consistent. Imagine that the inertial frame S
0
contains a single particle executing
8
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 5: The Earth! Shown is the polar axis (parallel to the angular velocity vector ), lines of
constant latitude (red), and lines of constant longitude (black).
a circular orbit anticlockwise in the (x, y) plane with radius r. Given that Newtons second law
holds unmodied in this frame, we know that there must be a centripetal force acting towards
the origin of magnitude
2
r, where is the angular velocity of the particle. Now consider a
frame S
1
which is rotating at angular velocity about the z-axis. The particle is stationary in
such a frame, as the rotation of S
1
is precisely such as to keep up with the orbiting particle. This
is consistent with the fact that in the rotating frame, there is a centrifugal force of magnitude

2
r acting outwards, which exactly counteracts the centripetal force...!
Which of the forces - Coriolis or centrifugal - is the dominant eect? This depends on the
situation - specically on the angular velocity of the rotating frame, the velocity of the particle
and also the position with respect to the rotation axis (both forces, however, are proportional
to the mass so that this makes no dierence to their relative size). A particle at rest in the
rotating frame, for example, experiences no Coriolis force. A particle moving on the surface of
the Earth, on the other hand, typically has a stronger Coriolis eect than a centrifugal eect, as
the angular velocity is so small (recall the Coriolis force is linear in , whereas the centrifugal
force is quadratic). We focus in more detail on motion near the Earths surface in the next
section.
2.4 Motion near the Earths surface
In order to discuss the surface of the Earth, it is convenient to introduce some standard ter-
minology. As shown in gure 5, the Earth rotates about the polar axis (i.e. the axis which
connects the North and South poles), with angular velocity . Two coordinates are conven-
tionally used to denote position on the spherical surface of the Earth
1
. These are the famous
latitude and longitude. Longitude measures how far around the Earth you are, azimuthally
with respect to the polar axis. It increases in an easterly direction, and is dened to be zero
on the Greenwich Meridian i.e. a line of constant longitude which passes through the Royal
Observatory in Greenwich, London. Latitude measures how far north or south you are, and is
dened to be zero on the equator, and increasing towards the north. Much of the confusion
about latitude and longitude for physicists who are not necessarily used to maritime navigation
(?!) results from the fact that they are not the usual angles we would use to describe a spherical
coordinate system. We would normally tend to use the polar angle, dened as increasing from
1
The Earth is not actually a perfect sphere. However, we will assume throughout this course that it is.
9
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 6: A cross-section through the Earth, rotated so that the polar axis is pointing upwards.
Shown are the North and South poles, and the denition of the latitude of a point on the Earths
surface. The vertical direction is shown, together with the direction of increasing latitude. The
direction of increasing longitude points into the page.
the polar axis, and the azimuthal angle in the xy plane. The latter is indeed the longitude, if
we take the xy plane to be the equatorial plane, with the x axis poking out through Greenwich.
However, latitude is measured from the equatorial plane, rather than from the polar axis. To
clarify this, we show a cross-section through the Earth in gure 6. The horizontal line denotes
the position of the equator, and latitude is dened as increasing in a northerly direction as shown.
A given point near the Earths surface will be specied by three coordinates: the latitude
and longitude, together with the vertical distance from the Earths surface. In gure 6 we
consider an example point on the Earths surface and show, for this point, the directions of
increasing vertical coordinate (labelled vert in the gure) and increasing latitude (labelled lat).
The direction of increasing longitude points into the paper at the point in question. A particles
velocity v can be specied by components in these three directions i.e. as a vector
v =
_
_
v
long
v
lat
v
vert
_
_
. (22)
Note that: (a) the coordinates are orthogonal at each point; (b) the ordering of the coordinates
in the above vector constitutes a right-handed system. The latter point is important if we are
to get minus signs correct in our equations, which will involve cross-products! Using a right-
handed coordinate system ensures that we can take cross products using the right-hand rule, as
is conventional.
The coordinates we have chosen on the surface of the Earth are in a rotating frame, and
thus we expect an object near the Earths surface to feel a Coriolis force
f
Cor.
= 2mv. (23)
In order to evaluate the above cross-product, we need to evaluate the Earths angular velocity
in the (long, lat, vert) coordinate system of eq. (22). By resolving components using the
10
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
geometry of gure 6, we nd
=
_
_
0
cos
sin
_
_
, (24)
where = || is the Earths angular speed. Thus, the Coriolis force on a particle whose velocity
is given by eq. (22) is
f
Cor.
= 2m
_
_
0
cos
sin
_
_

_
_
v
long
v
lat
v
vert
_
_
= 2m
_
_
v
vert
cos v
lat
sin
v
long
sin
v
long
cos
_
_
. (25)
This simplies in particular if we can neglect the vertical motion. That is, if we set v
vert
= 0
and ignore the vertical component of the Coriolis force. Then eq. (25) can be rewritten as
2m
_
_
v
lat
sin
v
long
sin
0
_
_
= 2m
_
_
0
0
sin
_
_

_
_
v
long
v
lat
0
_
_
. (26)
Or:
f
Cor.
= 2m
vert
v
where

vert
=
_
_
0
0
sin
_
_
(27)
is the projection of the Earths angular velocity in the vertical direction (i.e. upwards from
the Earths surface). Note that
vert
points in dierent directions in the northern and southern
hemispheres! In the northern hemisphere, the Earths rotation axis gets resolved onto the
upwards direction, and the Coriolis force points to the right with respect to the direction of
motion of an object. In the southern hemisphere, however, the angular velocity gets resolved
onto the downwards direction, and the Coriolis force then points to the left with respect to the
direction of motion of an object. We examine some of the consequences of this on the problem
sheet.
2.5 The Foucalt pendulum
As an example of motion near the Earths surface, we here consider the Foucalt Pendulum.
Named after the French physicist Leon Foucalt, this apparatus was exhibited in 1851 and pro-
vides a remarkably simple demonstration of the rotation of the Earth. Although there was also
more indirect evidence for the Earths rotation by this point in history (e.g. the equatorial
bulge, astronomical observations), Foucalts experiment provided a direct verication in a sim-
ple experiment which could be widely appreciated by both scientists and the general public.
There are many Foucalt pendula around the world, including one in Princes Square shopping
centre in Glasgow.
As its name suggests, the Foucalt pendulum is a large pendulum bob suspended somewhere
on the Earths surface. When left to oscillate, one nds that the plane of oscillation slowly
rotates about the vertical axis (see gure 7). The pendulum has to be fairly large (with a
heavy bob) to prevent air resistance from fully damping the motion before the rotation has
been observed.
We can understand the rotation of the plane of motion as follows. Let S
1
be the (rotat-
ing) frame of the Earth, and S
2
a frame rotating with angular velocity
2

vert
relative
11
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 7: A Foucalt pendulum is a large pendulum placed on the surface of the Earth. The plane
of motion of the pendulum is observed to slowly rotate about the vertical direction as shown.
to the Earth frame. By the acceleration transformation equation we derived previously, the
accelerations of the pendulum bob in the frames S
1
and S
2
satisfy
a
1
= R[a
2
+ 2
2
v
2
+
2
(
2
x
2
)], (28)
where x
i
, v
i
and a
i
are the position, velocity and acceleration of the bob in frame i respectively,
and R the appropriate rotation matrix between the frames. Assuming that both angular veloc-
ities and
2
are small, one can neglect the centrifugal term in eq. (28) (i.e. this is quadratic
in the angular velocity) to give
a
1
= R[a
2
+ 2
2
v
2
]. (29)
However, we also know that
a
1
= g 2
vert
v
1
, (30)
where g is the acceleration due to gravity, as seen in S
1
. This follows from the fact that S
1
is a
rotating frame, thus the pendulum bob experiences the acceleration that would also act in an
inertial frame (due to gravity), plus a Coriolis acceleration term involving the relevant angular
velocity. Given that the bob basically undergoes horizontal motion, we can neglect the vertical
motion and use the result from the previous section i.e. that the Coriolis acceleration depends
only on
vert
.
Given that we are neglecting quadratic terms in the angular velocities, we may write
v
1
Rv
2
(31)
in eq. (30). That is, there would ordinarily be an additional term involving
2
v
1
from the
velocity transformation law between rotating frames, but we can ignore this due to the further
cross-product with
vert
in eq. (30). Combining eqs. (30) and (29), we then nd
g 2R[
vert
v
2
] = R[a
2
+ 2
2
v
2
], (32)
where on the left-hand side we have used eq. (31) and pulled the rotation matrix R out of the
cross-product. We are allowed to do this, as we assumed above that the frame S
2
rotates about
12
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
the vertical axis i.e.
2

vert
. Then the rotation R commutes with the innitesimal rotation

vert
.
Looking at eq. (32) it is clear that if
2
=
vert
, the second terms on both the left- and
right-hand sides of eq. (32) cancel each other. Multiplying both sides by the inverse of the
rotation matrix, one is left with
a
2
= R
1
g. (33)
That is, the acceleration of the bob in the rotating frame S
2
is simply given by the acceleration
due to gravity, appropriately transformed.
The interpretation of the above result is as follows. In the frame S
2
, it looks as if only
gravity is present, so that the pendulum bob executes the normal oscillatory motion we expect
of a pendulum, with a well-dened plane of motion. Returning to the Earths frame S
1
, we
know that whatever happens in S
2
gets rotated about the vertical axis with angular speed
|
vert
| = sin. Thus, the plane of motion of the pendulum rotates around the vertical
direction, with an angular speed which depends on the latitude . For an alternative derivation
of this eect, and for some more discussion and historical context, there is a Wikipedia page:
http://en.wikipedia.org/wiki/Foucault pendulum
This concludes our discussion of rotating reference frames, which is only one example of
a non-inertial frame. However, we have nevertheless seen the main physical idea about non-
inertial frames - that of ctitious forces. These occur quite generally in accelerating frames,
with a familiar example being the force one feels on a rapidly decelerating train.
Next, we turn our attention to multiparticle systems.
3 Motion of a system of particles
So far, you will have extensively studied systems involving one or two particles, perhaps with
some forces acting. In general, mechanical systems contain many particles. Examples include:
atoms (which contain many protons, neutrons and electrons); gases, liquids and solids (which
contain many atoms or molecules); solar systems (containing a single star with many orbiting
asteroids and planets); galaxies of stars; clusters of galaxies. These examples highlight the fact
that many particle systems occur at many dierent length scales in the universe! The aim of
this section of the course is to study the classical mechanics of many particle systems, and to
put on a rigorous footing certain notions about how particles interact and behave.
In gure 8(a) we show a generic multiparticle system. There is indeed some cluster of par-
ticles, which may also have forces acting upon them, which may or may not be due to the
particles themselves. For example, charged particles will exert forces on each other due to their
electrostatic attraction or repulsion. Charged particles moving in a magnetic eld will experi-
ence an additional force, which is not due to the other particles but is instead due to the applied
magnetic eld.
In gure 8(b) we show our multiparticle system viewed from far away (i.e. from a distance
which is large compared to the separation of any two particles). If we are far enough away, we
perhaps expect the system to look eectively like a single particle, with some sort of average
motion associated with it. This particle will have an internal structure, due to the fact that it is
really made of lots of separate particles. However, we expect the average motion to decouple
in some well-dened sense from the internal motion of all the separate particles. We will shortly
13
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 8: A generic many particle system seen (a) at close proximity; (b) from far away. One
expects that, at large distances, the average motion of the system decouples from the internal
motion of the separate particles.
put this vague rambling on a more precise footing, but it is worth noting here some examples e.g.
atoms, galaxies of stars and even galactic clusters all look (from far away) like single objects,
with some internal structure. A more mundane, but no less beautiful, example is ocks of birds.
Seen from afar, there is an astonishing regularity in the motion of a ock, despite the fact that
it is composed of individual organisms.
In fact, we will see that one can indeed separate the motion of the centre of mass of a system
from the motion of particles relative to the centre of mass. The motion of the centre of mass
corresponds to the overall, or average motion of the system. The motion of particles relative
to the centre of mass corresponds to the internal motion of the constituents of the compound
many particle object, as seen from afar. Before making these ideas more concrete, let us rst
review some quantities that are used in describing the motion of single particles.
3.1 Single particles
A particle has a mass m, and a position (relative to the origin of the coordinate system used)
r. We can then dene the following quantities:
Velocity v = r
Momentum p = m r
Acceleration a = r
Angular Momentum L = r p
Kinetic Energy T =
1
2
m r
2
(34)
Note that the angular momentum is dened about a given point, namely the point from which
the vector distance r is measured. Note that, given we have used the position of the particle in
the above formula, we have taken the angular momentum about the origin.
A force F may be acting on the particle, and Newtons second law tells us that
F = ma =
dp
dt
. (35)
14
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
We may also dene the torque of the force F (taken about the origin)
G = r F. (36)
We can derive an equation analagous to Newtons second law, but expressed in ters of the torque.
Consider

L =
d
dt
(r p)
= r p +r p, (37)
where we have used the above denition of angular momentum about the origin. The rst term
on the right-hand side vanishes due to the fact that p r (i.e. so that r p = 0). We may use
Newtons second law in the second term to rewrite eq. (37) as
G =
dL
dt
, (38)
which is indeed very similar to Newtons second law: force is the rate of change of momentum;
torque is the rate of change of angular momentum. Having reviewed some aspects of single
particle motion, we now examine collections of many particles.
3.2 Many particles
We consider a system of N particles. In general, the masses may all be dierent, such that one
may label each particle by its mass m
i
and its position r
i
(from which one may also derive the
velocity and accleration). There may also be other parameters associated with each particle,
such as electric charge. However, we will not explicitly focus on such parameters in the subse-
quent discussion.
As in the single particle case, each particle may experience a force. In the mutliparticle
context, it is useful to separate the forces on each particle into two types:
1. External forces. These act on a single particle, such that particle i experiences a force
F
ext
i
, and result from some applied eld of force (e.g. a magnetic eld).
2. Internal forces. These act between pairs of particles, and arise due to the properties of
the particles themselves (e.g. electrostatic forces between charged particles; gravitational
attraction). We denote the force on particle i due to particle j as F
ji
.
This distinction is obviously absent in single particle systems, as then it is not possible to con-
sider a pair of particles which exert forces on one another.
We will see in what follows that Newtons third law plays a major role in separating the
overall motion of a system from its internal motion. In terms of the above notation, one has
F
ji
= F
ij
. (39)
That is, the force on particle i due to particle j is the opposite to that which particle j experiences
due to particle i. We will furthermore assume that the force that particles i and j experience acts
along the line joining i and j (gure 9). This is a very reasonable assumption (e.g. it is true for
electric and gravitational forces), and will be useful later when we consider angular momentum
properties of many particle systems. More mathematically, we may write this assumption as
F
ji
(r
i
r
j
), (40)
15
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 9: Diagram showing the force F
ji
on particle i due to particle j, and likewise for F
ij
.
Newtons third law gives F
ij
= F
ji
, and we also assume that the force acts along the line joining
the two particles.
as the right-hand side is a vector joining the two particles.
Having described a multiparticle system in general terms (also making explicit our assump-
tions on the internal forces), we examine the consequences of these assumptions in the following
sections, beginning with the notion of the centre of mass.
3.3 Centre of Mass
The total mass of our multiparticle system is given by
M =
N

i=1
m
i
, (41)
and the total momentum is obtained by summing over all individual momenta:
P =
N

i=1
p
i
=
N

i=1
m
i
r
i
. (42)
It is convenient to rewrite this in terms of the total mass M, which we can do by multiplying
and dividing by M to obtain
P = M
d
dt
_
i
m
i
r
i

i
m
i
_
= M

R, (43)
where in the second line we have introduced the centre of mass
R =
P
i
mi
ri
P
i
mi
Now consider the force on particle i. The total force will be given by the vector sum of the
external force acting on the particle, and all the internal forces (due to all the other particles).
More formally, we may write
F
i
= F
ext
i
+

j=i
F
ji
, (44)
16
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
where the sum is over all particles other than particle i. Newtons second law (F
i
= p
i
) gives
p
i
= F
ext
i
+

j=i
F
ji
, (45)
and from eq. (42) one nds

P =

i
_
_
F
ext
i
+

j=i
F
ji
_
_
. (46)
The second term involves the double sum
N

i=1

i=j
F
ji
=
N

i=1

j>i
F
ji
+
N

i=1

j<i
F
ji
, (47)
where on the right-hand side we have separated the terms which have j > i from those which
have j < i in the sum over j = i. Given that we are summing over both i and j, we can relabel
i j in the second double sum in eq. (47). We get
N

j=1

i<j
F
ij

N

i=1

j>i
F
ij
, (48)
where on the right-hand side we have interchanged the orders of the sums over i and j. This is
now almost the same as the rst term on the right-hand side of eq. (47), but with F
ji
F
ij
.
Plugging everything into eq. (46), this becomes
P =
N

i=1
F
ext
i
+
N

i=1

j>i
(F
ji
+F
ij
) . (49)
The second term, involving the sum over all internal forces, vanishes due to the fact that
F
ji
= F
ij
(Newtons third law). In words: we have summed over all the internal forces be-
tween all the particles. Every term involving the force on particle i due to particle j is cancelled
by a corresponding term involving the force on particle j due to particle i. The cancellation is
a consequence of Newtons third law.
Dening
F
ext
=
N

i=1
F
ext
i
(50)
to be the total external force acting on the system (i.e. the sum of all the individual external
forces), one nally nds

P = M
d
2
R
dt
2
= F
ext
. (51)
That is:
The centre of mass moves as a single eective particle of mass M under the total applied force.
A consequence of this statement is the following:
If there is no net applied force, the total linear momentum of a system of particles is conserved.
In this section, we have seen that Newtons second law holds for multiparticle systems con-
sidered as a single point mass with mass M (equal to the sum of the individual masses), and a
position at the centre of mass of the system. A similar statement holds for the rate of change
of angular momentum, which we consider in the following section.
17
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
3.4 Angular Momentum
From eq. (45) one has
m
i
r
i
= F
ext
i
+

j=i
F
ji
. (52)
The total angular momentum of our multiparticle system is given by
L =
N

i=1
r
i
p
i
=
N

i=1
m
i
r
i
r
i
. (53)
The rate of change of this is

L =
d
dt
N

i=1
m
i
r
i
r
i
=
N

i=1
m
i
d
dt
(r
i
r
i
) (54)
=
N

i=1
m
i
r
i
r
i
, (55)
where we have used the product rule in the last line, and also the fact that r
i
r
i
= 0. One
may substitute in the result of eq. (52) to get
N

i=1
m
i
r
i
r
i
=
N

i=1
r
i
F
ext
i
+
N

i=1

j=i
r
i
F
ji
. (56)
We may recognise the rst term on the right-hand side as
G
ext
=
N

i=1
G
ext
i
, G
ext
i
= r
i
F
ext
i
, (57)
i.e. the sum of the torques on each particle due to the external force each experiences. The
second term on the right-hand side of eq. (56) involves a double sum, and we may again use the
trick of separating out the j < i and j > i terms and relabelling i j to get
N

i=1

j=i
r
i
F
ji
=

j>i
(r
i
F
ji
+r
j
F
ij
)
=
N

i=1

j>i
(r
i
r
j
) F
ji
. (58)
As stated earlier, we have assumed that the internal force acting between particles i and j is
along the line joining them, as in gure 9. That is, we assumed F
ji
(r
i
r
j
). It then follows
that the contribution from eq. (58) vanishes, and we are left with

L = G
ext
, (59)
so that the rate of change of the total angular momentum of a multiparticle system is equal to
the total applied torque. That is, the system as a whole obeys the angular form of Newtons
second law. Note that a consequence of this is:
If no net torque is applied, the total angular momentum is conserved.
Again, we see that the system behaves as if it were a single particle with mass M and position
R (the centre of mass). We may now investigate further, and see exactly how the internal motion
separates from the centre of mass motion. This is the subject of the following two sections.
18
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
3.5 Separation of kinetic energy
The aim of this section is to show that, as far as kinetic energy is concerned, the internal motion
of a system is separate from the centre of mass motion. In order to do this, one may decompose
the position of particle i as follows:
r
i
= R+r

i
, (60)
which denes r

i
to be the (vector) displacement of particle i from the centre of mass. In other
words, r

i
is the position of particle i relative to the centre of mass. We will see that the total
kinetic energy and angular momentum of a multiparticle system can be written entirely in terms
of the centre of mass motion, plus terms which only depend on the relative displacements {r

i
}.
This is precisely what we mean when we say that the internal motion decouples from the centre
of mass motion.
It follows from the denition (60) that
N

i=1
m
i
r

i
=
N

i=1
m
i
r
i

i=1
m
i
R
= M
_
m
i
r
i

m
i
R
_
= 0, (61)
where we have used the denition of the centre of mass in the last line. We will use this result
several times in what follows.
Now consider the total kinetic energy T, which is given by
T =
1
2
N

i=1
m
i
r
2
i
=
1
2
N

i=1
m
i
_

R
2
+ 2 r

R+ ( r

i
)
2
_
=
1
2
N

i=1
m
i

R
2
+
1
2
N

i=1
m
i
( r

i
)
2
+

R
N

i=1
m
i
r

i
, (62)
where we have used eq. (60) in the second line. The nal term in the last line vanishes by
eq. (61), so that one has
T =
1
2
N

i=1
m
i

R
2
+
1
2
N

i=1
m
i
( r

i
)
2
. (63)
We may interpret this as follows. The total kinetic energy is the sum of all the internal kinetic
energies (i.e. measured relative to the centre of mass), and a kinetic energy of an eective par-
ticle whose mass is the total mass of the system, and whose position is the centre of mass. Put
another way, the internal motion of a multiparticle system separates from the overall motion of
the system, which is represented by the centre of mass motion.
This is important in any multiparticle system in which one may want to consider only the
internal motion, without worrying about the overall motion of the individual particles. An ex-
ample is our solar system. When we analyse the motion of planets, we neglect the fact that the
entire solar system is itself hurtling through space (i.e. due to the fact that the sun is rotating
about the centre of the Milky Way). We are allowed to neglect this, because we know that the
19
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
centre of mass motion decouples from the internal motion of the planets within the solar system,
as guaranteed by the above result.
Having seen how this separation works for the kinetic energy, we can also perform a similar
analysis for the angular momentum. This is the subject of the following section.
3.6 Separation of Angular Momentum
The total angular momentum of our multiparticle system is
L =
N

i=1
r
i
p
i
=
N

i=1
m
i
r
i
r
i
. (64)
Substituting eq. (60) this becomes
L =
N

i=1
m
i
(R+r

i
) (

R+ r

i
)
= MR

R+
_
N

i=1
m
i
r

i
_

R +R
_
N

i=1
m
i
r

i
_
+
N

i=1
m
i
r

i
r

i
. (65)
The second and third terms vanish using eq. (61), and we are left with
L = MR

R+L
internal
. (66)
That is, the total angular momentum is the sum of a term representing the angular momentum
of an eective particle of mass M about the origin, and
L
internal
=
N

i=1
m
i
r

i
r

i
(67)
is the sum of the internal angular momenta of the particles, evaluated about the centre of mass.
We also showed earlier that the rate of change of the total angular momentum was given by the
total torque due to the external forces F
ext
i
. That is,

L =
N

i=1
r
i
F
ext
i
= R
N

i=1
F
ext
i
+
N

i=1
r

i
F
ext
i
. (68)
We recognise the rst term as the total torque due to applied forces, of an eective particle of
mass M at the centre of mass R. The second term consists of the sum of all torques of the
individual particles about the centre of mass, due to the external force on each particle. Thus,
even when forces are acting, there is a well-dened separation of internal angular momentum
from the overall angular momentum about the origin (due to the centre of mass motion).
There are many physical applications of this separation of angular momentum. One such
example is a moving atom. We usually calculate the electron probability density for an atom
in terms of the orbital angular momentum L. That is, the electrons can be in dierent states
which are labelled by L (amongst other quantum numbers). However, there is also an overall
angular momentum due to the fact that the whole atom may be moving around the origin. The
20
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
above results tell us that the internal angular momentum states of the atom are completely
independent of the centre of mass motion, so that the electron states for a moving atom are the
same as for a stationary one.
This completes our review of many particle systems. We have seen that, subject to reasonable
assumptions on the internal forces in a many particle system, one may separate the overall
motion of a system (i.e. motion of the centre of mass) from the internal motion (motion relative
to the centre of mass). Up to now, we have talked almost exclusively using the framework of
Newtons laws. The aim of the rest of the course is to introduce an alternative viewpoint, which
is more general and powerful. We begin this presentation in the next section.
4 The Lagrangian formalism
5 Lagrangian
As we remarked at the beginning of the course, Newtons laws are the standard way that we
rst learn about classical dynamics. They distil thousands of years of thought about dynamics
in just three principles, whose applications and consequences reach far and wide.
The main use of Newtons Laws is in deriving the equations of motion of a system of particles.
For a general multiparticle system, consisting of particles of mass m
i
and position x
i
, the
equations of motion are given by Newtons second law applied to each particle:
m
i
d
2
x
i
dt
2
= f
i
, (69)
where f
i
is the total force acting on particle i. These equations (subject to suitable boundary
conditions) allow us, in principle, to calculate the trajectories x
i
(t) of all the particles in the
system. In order to do this, we need to be able to work out the total force on each particle, and
also be able to solve the resulting equations.
Unfortunately, problems arise in numerous contexts in constructing and solving these equa-
tions. Some of these problems are as follows.
Newtons laws apply only in inertial frames. We have seen earlier in the course that in
non-inertial frames we have to work hard to derive the correction terms to Newtons second
law. It would be nice to be able to circumvent this diculty.
There may be complicated constraints on the system, so that not all of the coordinates
x
i
are independent (we will see some examples later). This can introduce a great deal of
technical complexity in solving the equations of motion.
We may not know the forces f
i
before we solve the problem. As an example, consider a
system of gas molecules in a box, as shown in gure 10. We know that if a molecule hits a
wall, it will bounce o. We also know how it bounces o (i.e. how the force is related to the
particles momentum as it hits the wall). However, in order to work out the momentum
when a particle hits a wall, we need to solve the equations of the motion of the system!
We are caught in a bizarre Catch 22, of needing to know the forces to write down the
equations of motion, but knowing to solve the equations of motion to know the forces...
Due to the above problems, it is useful to have an alternative method for deriving equations
of motion, that avoids us having to think about forces, and also allows us to easily deal with
constraints. This alternative method is called the Lagrangian method. Not only does it simplify
many calculations of mechanical systems, it is also a much more powerful and general way of
21
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 10: Some molecules of gas in a box. The walls exerts a force on the particles when they hit
them, but we need to solve for the trajectories of these particles before we can know what the force
is!
analysing dynamical systems. It also reveals entirely new conceptual ideas underpinning classical
(and in fact quantum) mechanics, which are not so easily seen in the Newtonian approach. In
order to start the path to the Lagrangian approach, we must rst discuss constraints in detail.
This is the subject of the next section.
5.1 Constraints
In general, particles may be constrained, meaning that their coordinates are restricted in some
way. This may also imply that not all of the coordinates are independent from each other. A
classical example is the motion of a pendulum, shown in gure 11. If we set up the equations
of motion in terms of the coordinates (x, y) of the pendulum bob, we must somehow also im-
plement the fact that these coordinates are not independent - they must satisfy the relation
x
2
+ y
2
= L
2
as shown in the gure, where L is the length of the pendulum.
In this case the constraint is actually fairly easy to deal with. Instead of using x and y, one
uses the angle of the pendulum string with the vertical to specify the position of the bob.
One may then parametrise x and y in terms of i.e. (up to arbitrary constants) x = Lsin,
y = Lcos . One then nds a single equation of motion for .
In general, we may think of constraints as being implemented by constraint forces. In the
pendulum example, the tension in the string acts as the constraint force, restricting the bob to
follow a circular path. The total force on a particle i in a system with constraints is
f
i
= F
appl.
i
+f
const.
i
, (70)
where f
const.
i
is the constraint force, and F
appl.
i
the applied force on a particle. In the pendulum
example, the applied force is that due to gravity (mg), whereas the constraint force is the ten-
sion in the string, which acts perpendicularly to the motion of the bob. The confusing nature
of constraint forces (in many cases, they seem somewhat weird or arbitrary, such as normal
contact forces) is one of the reasons why Newtons laws are undesirable - we will see later that
we can avoid thinking about them in the Lagrangian approach.
There are dierent types of constraint. An important class is that of holonomic constraints.
A holonomic constraint is dened to be any constraint that implies a constraint equation linking
22
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 11: A pendulum. The bob has x and y components, which are constrained by the fact they
must lie on a circle of radius L.
one or more of the particle coordinates. That is, one may write an equation of the form
g(x
1
, x
2
, . . . , x
N
, t) = 0 (71)
for some function g. Note that g may depend also on time in general. If so, it is called
rheonomous. If it does not depend on time (i.e. the constraint is time-independent), it is called
scleronomous. Examples of holonomic constraints include
1. The pendulum we looked at above, whose coordinates satisfy the constraint equation
x
2
+ y
2
L
2
= 0.
2. A particle xed to move along a curve (which may depend on time). The constraint
equations are then such as to specify the curve.
3. A particle moving on the Earths surface, whose coordinates (if we take Cartesian coor-
dinates instead of spherical polars) obey the constraint equation x
2
+ y
2
+ z
2
R
2
= 0,
where R is the radius of the Earth.
It may seem that all constraints are holonomic. However, this is not true! Examples of
non-holonomic constraints are:
1. A particle moving on the surface of a sphere, which may fall o the sphere. The coordinates
satisfy x
2
+ y
2
+ z
2
R
2
, which is not a constraint equation, as it involves an inequality.
2. Friction and air resistance forces frequently involve the velocity of a particle, rather than
its coordinates directly.
From now on, we only consider systems which have holonomic constraints
2
. For such systems,
we can always implement the constraint by transforming to a set of generalised coordinates,
constructed to be independent from each other. By this, I mean that we can nd a set of
independent quantities {q
k
} which we may parametrise the coordinates x
i
in terms of i.e.
x
i
= x
i
({q
j
}, t) (72)
2
The whole issue of dealing with non-holonomic systems is a complex one, and is still generating new research.
23
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 12: A double pendulum, for which the natural generalised coordinates are the two angles
shown.
for each of the particle coordinates x
i
. If we are in D dimensions (we may as well be com-
pletely general!), each position vector has D components. If there are N particles subject to
k constraints, the number of independent degrees of freedom (i.e. the number of generalised
coordinates we need) is
N
d.o.f.
= DN k. (73)
Note that a given set of generalised coordinates is not unique (in the same way that we may
choose dierent coordinate systems in unconstrained systems) , but the number of them is xed
by the number of constraints according to eq. (73).
An example will perhaps clarify the above discussion. Consider again the pendulum. This
lives in two dimensions (considering only the motion in the plane), and there is one constraint
equation. Then the number of generalised coordinates we need is DN k = 2 1 1 = 1. We
have already seen that this is indeed the case (i.e. we need a single angle ).
In general, the generalised coordinates will not be orthogonal, or even easily visualisable in
the same way that e.g. Cartesians are, in terms of well-dened coordinate axes. As an example,
consider the double pendulum of gure 12, consisting of a pendulum attached to the bob of
another pendulum. Here we need two generalised coordinates, and it is particularly convenient
to choose the two angles shown in the gure. These are not directly visualisable in terms of e.g.
orthogonal directions in space. Nevertheless, it is clear in this case what their physical meaning
is (angles). In more general cases, a simple physical interpretation of the generalised coordinates
may be unclear!
Finally, I note that the idea of generalised coordinates is useful even if no constraints are
present. For example, one may regard spherical polar coordinates in three dimensions as being
generalised coordinates taking the place of Cartesians in situations where this is more convenient
(e.g. a particle in a spherically symmetric potential).
24
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
5.2 DAlamberts Principle and Lagranges Equations
Having introduced generalised coordinates, we now proceed to derive the equations of motion
which are satised by these coordinates. Note that, becuase we have chosen the {q
i
} to be
independent, we do not need to worry about implementing additional constraints when solving
these equations.
The equations we will derive are known as Lagranges equations. There are two main ways
to derive them. One (dAlamberts principle) is more old-fashioned and historical, and involves
a much longer argument. However, it also makes the assumptions about the physics absolutely
clear. The second method (the principle of least action) is more elegant and quick, but more
mysterious and less obviously physical at rst sight. Thus, we will follow dAlamberts approach
to begin with, and later on consider the alternative.
Let us begin with Newtons second law f
i
= p
i
, which can be rewritten as
f
i
p
i
= 0. (74)
Now consider a virtual displacement of the system. That is, we imagine that each of the particles
(with position r
i
) gets displaced by a small amount r
i
, where the displacements satisfy all of
the constraints which may be acting. From eq. (74), we clearly have

i
(f
i
p
i
) r
i
= 0. (75)
This is dAlamberts principle. Admittedly in this context it looks like an arbitrary restatement
of Newtons second law. However, this expression is of historical importance, and also will lead
us to Lagranges equations, which we will see are more powerful than Newtons laws.
DAlamberts principle is a consistency relation between the total forces acting on particles,
their momenta, and any set of virtual displacements consistent with the constraints of the
system. To work out the further consequences of eq. (75), one may decompose the total force f
i
on each particle as the sum of the constraint force and the applied force, as in eq. (70), to get

i
_
F
appl.
i
p
i
_
r
i
+

i
f
const.
i
r
i
. (76)
We now assume that the constraint forces do no work. One way this can happen, for example,
is if they are perpendicular to the motion of the particles. As an example, consider the pendu-
lum of gure 11. The constraint force in this case is the tension of the string, which is indeed
perpendicular to the motion of the particle. Other examples include the normal contact force
of particles on a surface, or friction for a car which goes round a roundabout (i.e. the friction
force is centripetal, acting towards the centre of the circle).
If the constraint forces do no work, the second sum in eq. (76) vanishes and we are left with

i
_
F
appl.
i
p
i
_
r
i
= 0, (77)
from which it does not follow that F
appl.
i
= p
i
or each particle i!! This is because the dis-
placements r
i
are not independent in general. Thus, if we vary one of them then another may
change, such that cancellations may occur between terms involving dierent displacements.
25
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
We saw above, however, that we can implement constraints by choosing a set of generalised
coordinates {q
i
} which are indeed independent. Thus, if we rewrite eq. (77) in terms of varia-
tions of the generalised coordinates, q
i
, rather than the r
i
, we can nd a set of equations of
motion based on the fact that the q
i
can be varied independently from each other.
To do this, recall that the coordinates r
i
are parameterised in terms of the generalised
coordinates via r
i
= r
i
({q
i
}, t). Using the chain rule, one then nds
r
i
=

j
r
i
q
j
q
j
. (78)
Similarly, velocities may be written as
v
i
=
dr
i
dt
=

k
r
i
q
k
q
k
+
r
i
dt
, (79)
where again the chain rule has been used. From this, we nd that
r
i
q
j
=
v
i
q
i
, (80)
which we will use later on. From eq. (78), the rst term of eq. (77) can be rewritten as

j
F
appl.
i

r
i
q
j
q
j
=

j
Q
j
q
j
, (81)
where we dened the generalised force
Q
j
=

i
F
i

r
i
q
j
. (82)
We also need to rewrite the second term of eq. (77) i.e.

i
p
i
r
i
=

j
m
i
r
i

r
i
q
j
q
j
. (83)
First note that, by using the product rule, one may write
d
dt
_
m
i
r
i

r
i
q
j
_
= m
i
r
i

r
i
q
j
+ m
i
r
i

d
dt
_
r
i
q
j
_
. (84)
Thus, one may write

i
m
i
r
i

r
i
q
j
=

i
_
d
dt
_
m
i
r
i

r
i
q
j
_
m
i
r
i

d
dt
_
r
i
q
j
__
. (85)
Here the last term involves
d
dt
_
r
i
q
j
_
=

k
q
k

q
k
_
r
i
q
j
_
+

t
_
r
i
q
j
_
=

2
r
i
q
j
q
k
q
k
+

2
r
i
q
j
t
=

q
j
_

k
r
i
q
k
q
k
+
r
i
t
_
=
v
i
q
j
.
26
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
We have done a number of things here. In the rst line, we used the chain rule to rewrite the
total derivative with respect to t in terms of partial derivatives of q
k
and t. Then we pulled
out the partial derivative with respect to q
j
, and recognised eq. (79) in the last line. We may
substitute this result into eq. (85), and also use eq. (80) in the rst term on the right-hand side
to get

i
m
i
r
i

r
i
q
j
=

i
_
d
dt
_
m
i
v
i

v
i
q
j
_
m
i
v
i

v
i
q
j
_
. (86)
Noting that
m
i
v
i

v
i
q
j
=

q
j
_
1
2
m
i
v
i
v
i
_
, m
i
v
i

v
i
q
j
=

q
j
_
1
2
m
i
v
i
v
i
_
, (87)
and recognising
T =

i
1
2
m
i
v
i
v
i
(88)
as the total kinetic energy of the system, one nally nds

i
p
i
r
i
=

j
_
d
dt
_
T
q
j
_

T
q
j
_
q
j
. (89)
We have gradually been rewriting dAlamberts principle. Collecting everything weve done, we
can substitute eqs. (81) and (89) into eq. (77) to get

j
_
d
dt
_
T
q
j
_

T
q
j
Q
j
_
q
j
= 0, (90)
which relates the generalised forces Q
j
to derivatives of the kinetic energy. We have now achieved
what we set out to do, which is to rewrite dAlamberts principle in terms of displacements in the
generalised coordinates. Given that these coordinates are all mutually independent, we can set
the coecient of each term q
j
in eq. (90) separately to zero. We then get the set of equations:
d
dt
_
T
q
j
_

T
q
j
= Q
j
. (91)
These are in principle the equations of motion of our system. They still look rather abstract,
though, particularly as they involve the generalised forces Q
j
dened by eq. (82). However, we
may usually assume that the applied forces F
appl.
i
can be derived from a potential energy i.e.
F
appl.
i
=
i
V (r
i
), (92)
where
i
= /x
i
. Then the generalised forces are given by
Q
j
=

i
F
appl.
i

r
i
q
j
=

i
V
r
i
q
j
=
V
q
j
,
where we have used the chain rule in the last line. Substituting this in eq. (91) gives
d
dt
_
T
q
j
_

(T V )
q
j
= 0. (93)
27
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Finally, let us assume that the potential energy V does not depend on the generalised velocities
{ q
i
} i.e.
V
q
i
= 0. (94)
This allows us to replace T by T V in the rst term of eq. (93), so that it becomes
d
dt
_
L
q
j
_
=
L
q
j
, (95)
where we dened the Lagrangian
L = T V. (96)
Eqs. (95) are Lagranges equations
3
. They are the equations of motion of our system, where
there is one equation of motion for each generalised coordinate q
i
. These equations can be
regarded as an alternative to Newtons Laws. We can see at rst sight why they are easier to
apply in general - we only need to worry about the potential energy and kinetic energy, which
are usually easier to work out than the forces which are acting on a system (particularly when
constraints are involved).
We will see some examples of the above equations shortly, and indeed the rest of the course
will be concerned with these equations in various contexts, and also the consequences of them
and what they reveal about classical dynamics. Thus, Lagranges equations are highly important
and you should remember them! It is perhaps not so important to be able to derive them from
dAlamberts principle, but given the very technical discussion above it is perhaps useful to
summarise all the assumptions we made in deriving the equations. We assumed the following:
1. The system of interest has holonomic constraints i.e. those which can be written as in
eq. (71).
2. The constraint forces do no work.
3. The forces are derived as the gradient of a potential function which does not depend upon
velocities.
4. The generalised coordinates {q
i
} are independent (in fact, this is part of their denition
i.e. we can always choose this).
5. The generalised coordinates {q
i
} and the generalised velocities { q
i
} can be treated as
independent variables. This was used above when we used the chain rule at some point.
In this course, we are not going to consider systems where the above assumptions are not valid,
but it is worth merely noting here that things are complicated in such cases...!
Another thing worth noting is that the denition of the Lagrangian L is not unique. One
can redene it according to the transformation
L

= L +
dF({q
i
}, t)
dt
(97)
(where F is an arbitrary dierentiable function), which leads to the same equations of motion
(see the problem sheet).
Let us now consider some examples of some simple systems, and their corresponding Lagrange
equations.
3
Note they are also sometimes referred to as the Euler-Lagrange equations.
28
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
5.2.1 Example 1: Particle in one dimension
Almost the simplest dynamical system you can get is a particle moving in one dimension x, with
mass m and potential energy V (x). In this case there are no constraints, and we can take the
variable x to be our only generalised coordinate. The Lagrangian is given by
L = T V =
1
2
m x
2
V (x). (98)
Then we have
L
x
= m x,
L
x
=
V
x
=
dV
dx
, (99)
where in the last step we used the fact that the potential only depends on x, so that we
can replace the partial derivative with an ordinary derivative. Plugging these results into the
Lagrange equation for x (eq. (95)) gives the equation of motion
m x =
dV
dx
. (100)
We recognise this as being the appropriate form of Newtons second law for this system. Thus,
the equation of motion we get from the Lagrangian approach is the same as what we get from
Newtons laws. The correspondence is particularly simple here, as there are no constraints
acting. However, it must be true in general that the equations of motion from both approaches
explain the same dynamics i.e. are equivalent physical descriptions.
5.2.2 Example 2: The pendulum (again!)
We kept mentioning the pendulum above, mainly because weve all studied it for years, so its
a good example to use for clarifying all the new concepts weve introduced above.
As we have already remarked, there is a single generalised coordinate for the pendulum,
which is the angle shown in gure 11. The kinetic and potential energies for the pendulum
bob are
4
T =
1
2
l
2

2
, V = mlg cos . (101)
respectively, so that the Lagrangian is
L =
1
2
ml
2

2
+ mlg cos . (102)
Then we have
L

,
L

= mlg sin. (103)

The Lagrange equaton for is then
ml
2

= mlg sin , (104)

or

=
g
l
sin . (105)
Usually one is interested in simple pendula, i.e. those for which we know the angle is small.
Then we can set sin in eq. (105) and get the simple harmonic motion equation

=
g
l
. (106)
4
Note that I will now use l to denote the pendulum length rather than L, so as not to confuse notation with the
Lagrangian.
29
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 13: A bead on a rotating wire of constant angular velocity . (a) Sideways view; (b) Looking
down the rotation axis.
We could also have gotten this equation from Newtons laws of course, but the above derivation
is easier (in my opinion at least). In particular, we dont have to fa about with tension forces
etc.
5.2.3 Example 3: Bead on a rotating wire
As a nal example for now, consider a wire which is perpendicular to a vertical axis, and rotates
about this axis. Let there be a bead on this wire which is free to move along it, as shown in
gure 13. What happens as the wire spins with constant angular velocity? It is interesting rst
to note that this is an example of a system with a time-dependent constraint function. That is,
referring to gure 13(b) one has
_
x
y
_
=
_
r cos t
r sin t
_
, (107)
where is the angular speed, so that the constraint equation is
arctan
_
y
x
_
t = 0. (108)
The most natural generalised coordinate to use is r, the distance of the bead along the wire, as
shown in gure 13(b). Note that this is a coordinate in a rotating frame. However, we dont
have to worry about this as much as we did in the Newtonian framework! We just write down
the kinetic and potential energy. There is no potential energy in this case, and the kinetic energy
is
T =
1
2
m
_
r
2
+
2
r
2
_
= L, (109)
where the last equality follows from L = T V . One then has
L
r
= m r,
L
r
= m
2
r, (110)
so that Lagranges equation for r is
r =
2
r. (111)
The solution of this equation is in fact
r = Ae
t
+ Be
t
, (112)
30
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Figure 14: Some possible trajectories with xed endpoints q
i
(t
0
) and q
i
(t
1
).
where A and B are constants which depend upon the boundary conditions. For large times this
behaves like r e
t
, and thus we see that the bead is ung exponentially outwards. This makes
sense if we think about what happens if we analyse the bead in the Newtonian framework. As
remarked above, r is a coordinate in a rotating reference frame. Thus the bead sees a centrifugal
force acting outwards from the rotation axis. There is nothing to balance this force, so that
the bead gets ung outwards as shown in the Lagrangian approach. You will presumably have
veried this phenomenon at some point in your life, by whirling something round your head and
letting it go.
5.3 Principle of Least Action
In the previous section, we found a general prescription for deriving equations in motion in
arbitrary coordinate systems. This derivation was quite long and involved, and the aim of this
section is to show how Lagranges equations can be derived more elegantly. This derivation is a
little more mysterious however, which is why we presented dAlamberts approach rst.
Given a Lagrangian L, we can dene a quantity called the action via
S[{q
i
}, { q
i
}] =
_
t1
t0
dtL({q
i
}, { q
i
}, t). (113)
That is, the action is the time integral of the Lagrangian between some initial time t
0
and some
nal time t
1
> t
0
. As denoted on the left-hand side, the action depends on the trajectories
{q
i
(t)} (and their time derivatives). The mathematical terminology is that S is a functional
of each q
i
(t). A function takes a number and returns a number, whereas a functional takes a
function (of t in this case) and returns a number.
Let q
i
(t
0
) and q
i
(t
1
) be xed for each q
i
i.e. each particle trajectory is xed at the initial and
nal times. Then there are an innite number of possible trajectories which join these two xed
end points, as shown in gure 14. The particle follows only one of these trajectories, namely
the particular trajectory which satises the equations of motion. There is also another way to
say which trajectory is chosen, which is the following:
The particle follows the trajectory that extremises the action S.
31
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
This is known as principle of least action, where the name reects the fact that the extremum
of the action is usually a minimum. To show that this is correct, we must show the equations
of motion (Lagranges equations) are correctly reproduced by requiring that S be extremised.
To do this, assume that each q
i
is indeed such that S is extremised. Then one may introduce
a small perturbation to each q
i
:
q
i
(t) q
i
(t) + q
i
(t), (114)
where q
i
(t) is dened at every point along the trajectory i.e. at all times t. Note that, due
to the xed end points of the trajectory, q
i
(t
0
) = q
i
(t
1
) = 0. It follows that the generalised
velocities get perturbed according to
q
i
q
i
+ q
i
, q
i
=
d
dt
q
i
. (115)
The action will also be perturbed, that is
S S + S =
_
t1
t0
dtL(q
i
+ q
i
, q
i
+ q
i
, t), (116)
where on the right-hand side we have just substituted eqs. (114) and (115) into the denition
of the action, eq. (113). Using Taylors theorem, we can rewrite this as
S + S =
_
t1
t0
dt
_
L(q
i
, q
i
, t) +

i
_
L(q
i
, q
i
, t)
q
i
q
i
+
L(q
i
, q
i
, t)
q
i
q
i
_
_
+ . . . , (117)
were we neglect terms which are quadratic or more in the perturbations. The rst term of
eq. (117) is just the original action S. The remaining terms are
S =

i
_
t1
t0
dt
_
L
q
i
q
i
+
L
q
i
q
i
_
. (118)
Using that q
i
= d(q
i
)/dt, one may integrate by parts in the second term to get
S =

i
_
_
L
q
i
q
i
_
t1
t0
+
_
t1
t0
dt
_
L
q
i

d
dt
_
L
q
i
__
q
i
_
. (119)
The rst term in the curly brackets vanishes due to the fact that the trajectory has xed
endpoints (q
i
= 0 at t = t
0
and t = t
1
). Furthermore, if S is extremised, then one must have
S = 0 given that we assumed we are perturbing about the trajectory which indeed extremises
S. Thus one has

i
_
t1
t0
dt
_
L
q
i

d
dt
_
L
q
i
__
q
i
= 0. (120)
Crucially, we have not specied what the perturbations q
i
are i.e. this equation must be satised
for any choice of small perturbations. Also, the generalised coordinates are independent of each
other, therefore so are their variations. Thus, the only way for eq. (120) to be satised is if the
contents of the square bracket vanishes for each q
i
. In other words one has
d
dt
_
L
q
i
_
=
L
q
i
, (121)
which are precisely Lagranges equations! We see that the principle of least action indeed works
in this case, reproducing the known physics of dynamical systems that we derived in a more
32
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
physical way from dAlamberts principle.
The reason for presenting this derivation second is that at rst sight its a lot more myste-
rious. Where, for example, does the principle of least action come from? Historically, it was
motivated by work in optics (i.e. the rst least action principle was used by Fermat to nd the
path that light rays follow in a medium of nontrivial refractive index). Nowadays, however, the
principle of least action has the status of a fundamental principle of nature. Asking where the
principle of least action comes from is akin to asking where Newtons laws come from. Both
are sets of assumptions, and both ultimately lead to the same physical description of dynamical
systems.
Many physicists regard the principle of least action as being in a sense more fundamental
than e.g. Newtons laws. The reason for this is that all theories in modern physics can be written
down using the principle of least action! Above, we saw what the Lagrangian looked like for a
system of non-relativistic particles. However, one can also write down Lagrangians which rep-
resent e.g. relativistic particles, electric and magnetic elds (leading to Maxwells equations),
gravitational elds (leading to general relativity)... There are also Lagrangians describing the
strong nuclear force (quantum chromodynamics) and the weak nuclear force (which gets unied
with the electromagnetic force). Furthermore, there is a well-dened prescription for taking a
classical theory dened by a Lagrangian, and turning it into a quantum theory. In the case of
particles, one gets quantum mechanics. In the case of elds, one gets quantum eld theory. In
the case of strings, one gets string theory, a very rich theory whose consequences are still being
explored. Thus, the principle of least action underpins all of physics, and there is currently no
reason to believe that any new theory that we could come up with would not be describable
within the Lagrangian formalism.
Having now thoroughly derived Lagranges equations, we examine some of their consequences
in the following sections.
5.4 Generalised momentum
Recall that the Lagrangian for a particle moving in one dimension is
L =
1
2
m x
2
V (x), (122)
from which we nd that
L
x
= m x. (123)
We recognise the right-hand side as the linear momentum of the particle. This leads us to dene,
for some general Lagrangian L({q
i
}, { q
i
}, t), the generalised momentum
p
i
=
L
q
i
. (124)
In the jargon of classical dynamics, we call p
i
the canonical momentum conjugate to q
i
, or just
the conjugate momentum. At this stage, one should just think of this as some quantity that
we dene. It is sometimes, but not always, equal to the linear momentum of a particle! For
example, the Lagrangian for the pendulum of gure 11 was
L =
1
2
ml
2

2
mgl cos , (125)
which gives the canonical momentum (conjugate to )
p

=
L

= ml
2

. (126)
33
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
We recognise this in fact as the angular momentum of the pendulum bob (about the point at
which the pendulum is xed). It is often true that momenta conjugate to angles are standard
angular momenta, whereas those conjugate to linear coordinates are linear momenta. However,
in general some coordinate q
i
may not have a simple physical interpretation, so that the conju-
gate momentum will be some more abstract quantity that we have not encountered before.
The notion of conjugate momenta leads to the idea of conservation laws. We explore this
further in the next section.
5.5 Symmetries and Conservation Laws
In the previous section, we introduced canonical momenta conjugate to the generalised cood-
inates {q
i
}. In this section, we see that symmetries of a system lead to conservation laws
regarding these momenta.
Consider the case where a Lagrangian L does not depend explicitly on one of the generalised
coordinates q
i
. Then we have
L
q
i
= 0. (127)
It immediately follows from the Lagrange equation for q
i
that
d
dt
_
L
q
i
_
= p
i
= 0, (128)
i.e. that the canonical momentum conjugate to q
i
is conserved. This is a very beautiful result.
Most systems in physics indeed possess some symmetry e.g. translation invariance, rotational
invariance etc. The conservation of momentum and angular momentum, which we learn about
early on in the Newtonian framework, is purely a consequence of the fact the a system possesses
such a symmetry. In a sense, we have derived the conservation of momentum from a more
fundamental idea - that of symmetry. The whole idea of symmetries being related to conservation
laws is used to great eect in e.g. the Standard Model of particle physics. There, the symmetries
of the elds describing various particles get mathematically rather abstract... Let us look at
some more everyday examples.
1. Free particle in 3 dimensions. This has Lagrangian
L =
1
2
m x x. (129)
Thus, L does not depend on x
i
(only through the velocity), and so the linear momentum
p = m x is conserved.
2. System of particles in 3D. Consider a multiparticle system, with no applied force. The
forces acting on a given particle are then only due to the other particles, and we may write
the Lagrangian for the system as
L =

i
1
2
m
i
x
i
x
i
V ({x
i
x
j
}), (130)
where the potential energy explicitly depends only on the relative displacements of the
particles. As we saw in eq. (60), we can decompose the position of a particle in terms of
the displacement relative to the centre of mass R. In the present notation we have
x
i
= R+x

i
, (131)
34
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
so that the Lagrangian of eq. (130) becomes
L =
1
2
M

R
2
+

i
1
2
m
i
x

i
x

i
V ({x

i
x

j
}), (132)
where we have used the fact that x

i
x

j
= x
i
x
j
, and M is the total mass of the system.
We see that the Lagrangian does not explicitly depend on the centre of mass R (i.e. only
on

R), so that the total linear momentum M

R is conserved. We saw already in section 3
that total linear momentum was conserved if there is no applied force. Here we rederive
this as a consequence of transational invariance.
5.6 The Energy Function (Hamiltonian)
It is useful to consider the derivative of L with respect to time. Using the chain rule, and the
fact that L is manifestly a function of q
i
, q
i
and t, one nds
dL
dt
=

i
L
q
i
dq
i
dt
+

i
L
q
i
d q
i
dt
+
L
t
. (133)
Using Lagranges equations, we can rewrite the rst term to get
dL
dt
=

i
_
d
dt
_
L
q
i
_
dq
i
dt
+
L
q
i
d q
i
dt
_
+
L
t
=

i
d
dt
_
L
q
i
q
i
_
+
L
t
. (134)
It thus follows that
dH
dt
=
L
t
, (135)
where
H =

i
q
i
L
q
i
L (136)
is called the Hamiltonian, named after the Irish mathematician William Hamilton. Another
name for this function is the energy function, and we can see how it gets this name by considering
e.g. a particle in one dimension with Lagrangian
L =
1
2
m x
2
V (x). (137)
Applying eq. (136), one nds a Hamiltonian
H = x(m x)
1
2
m x
2
+ V (x)
=
1
2
m x
2
+ V (x)
= T + V = E (138)
where E is the total energy. It is quite generally true that the Hamiltonian represents the total
energy of the system (you may have seen the Hamiltonian operator being used in quantum
mechanics).
Note from eq. (135), we have that total energy is conserved if the Lagrangian does not ex-
plicitly depend upon time. This is an analogue of the conservation of canonical momentum,
35
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
with a slight caveat. Time and energy are not quite conjugate variables in the same way that
generalised coordinates and their canonical momenta are. That is, time is not a generalised co-
ordinate. This distinction may seem somewhat pedantic, but there are a number of conceptual
issues that arise in e.g. quantum mechanics, in which it is easy to get confused if one forgets
that energy and time are not quite on the same footing as position and momentum.
This is all Im going to say about the Hamiltonian in this course. There is actually a lot
more that can be done here. In particular, one can set up classical dynamics purely in terms
of the Hamiltonian, rather than the Lagrangian. The equivalent of Lagranges equations are
called Hamiltons equations, and they express the time derivatives of the generalised coordinates
and canonical momenta in terms of the Hamiltonian function. This formalism is particularly
well-suited to examine the relationship between classical and quantum mechanics for example.
In this section of the course, we have introduced the Lagrangian formalism for deriving equa-
tions of motion. Instead of worrying about forces, one only needs to construct the Lagrangian
function, which is the dierence between the total kinetic and potential energies. Once one has
L, there is a well-dened recipe for obtaining equations of motion, and this procedure works for
any choice of generalised coordinates, thus is not restricted to particular frames of reference.
In this sense, the Lagrangian formalism is generally much easier to apply than Newtons laws,
especially for more complicated systems that involve the use of non-inertial frames. We have
also seen that Lagrangian mechanics reveals powerful aspects of classical dynamics that are hid-
den in the Newtonian approach, such as the relationship between symmetries and conservation
laws. In the nal part of the course, we consider a particular example of the application of the
Lagrangian formalism, namely the theory of small oscillations. This is the subject of the next
section.
6 Small oscillations
If you kick an inanimate object, it usually wobbles slightly and then comes to rest. The latter
eect is due to friction forces, but the wobbling is a consequence of the fact that all dynamical
systems, if displaced about a stable equilibrium position, undergo small oscillations. The aim
of this section is to investigate this phenomenon. The analysis of oscillations is not just limited
to classical dynamical systems, but also has applications throughout other areas of physics and
chemistry (e.g. energy spectra of molecules are related to the various ways in which they can
vibrate).
To illustrate the idea of small oscillations, we consider a particular example (the double
pendulum) before discussing the general framework.
6.1 The double pendulum
Consider the double pendulum of gure 12, consisting of a pendulum attached the bob of a
second pendulum. We assume for simplicity that the two pendula have the same length a, and
also that the bobs have the same mass m. The potential energy of the system is
V = mga cos mga(cos + cos ), (139)
and the kinetic energy is
T =
1
2
ma
2

2
+
1
2
ma
2
(

+

)
2
, (140)
where care is needed for the second pendulum bob i.e. its angular velocity about the point at
which the entire system is xed is given by the sum of the angular speeds of the rst bob, and
36
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
that of the second bob relative to the rst pendulum bob. One then squares this combination
to make the kinetic energy.
Given the kinetic and potential energies, we can then form the Lagrangian
L =
1
2
ma
2

2
+
1
2
ma
2
(

2
+

2
+ 2

) + mga(2 cos + cos )

1
2
ma
2

2
+
1
2
ma
2
(

2
+

2
+ 2

) mga
_

2
+
1
2

2
_
(141)
where in the second line we have assumed that both angles are small, so that we can expand
the cosines according to
cos 1
1
2

2
+ . . . . (142)
The equations of motion are given by Lagranges equations, which turn out to be
2

+

+
2g
a
= 0 (143)

+

+
g
a
= 0. (144)
This is a set of coupled second-order ordinary dierential equations. There are two equations,
which when solved give the two functions (t) and (t). Each equation looks a bit like the
well-known simple harmonic oscillator equation:
x +
2
x = 0, (145)
where is the angular frequency of the oscillator. Indeed, the solutions of eqs. (143) and (144)
are combinations of and which are oscillating with a single frequency. To show this, consider
the trial solution
_

_
=
_
c

e
it
c

e
it
_
, (146)
where c

and c

are constants, which may be complex in general (i.e. corresponding to some

initial phase of the oscillation). This solution is just what we mean by the two coordinates
oscillating with a single frequency . We will now substitute this form into the equations of
motion, and nd that there are two such solutions.
From eq. (146) one nds

=
2
,

=
2
. (147)
Plugging these into eqs. (143,144), one gets the relations
_
2
2
+
2g
a
_
c

2
c

= 0

2
c

+
_

2
+
g
a
_
c

= 0.. (148)
Note that we can write these in matrix form as
_
2g/a 2
2

2
g/a
2
__
c

_
= 0. (149)
Our trial solution thus indeed ts, provided that eq. (149) is satised. This in turn implies that
the determinant of the matrix vanishes i.e.

2g/a 2
2

2
g/a
2

= 2
_
g
a

2
_
2

4
= 0. (150)
37
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
This is a quadratic equation for
2
, and has two solutions:

2
=
g
a
(2

2). (151)
There are thus two solutions of the form of eq. (146), consisting of combinations of the coordi-
nates (specied by the constants c
i
) which oscillate with a single frequency. The frequencies
are called normal frequencies of the system, and the combinations of coordinates (given by the
vector of constants c
i
) are referred to as normal modes.
A normal mode is some collective motion of a multiparticle system, which has a single oscil-
latory frequency associated with it. Each normal mode is specied by this frequency, together
with the vector of constants c
i
. To nd the normal modes for the double pendulum example, we
have to substitute each normal frequency in turn into the matrix relation of eq. (149). Taking
rst
2
= g(2 +

2)/2a we get
g
a
_
2 4 2

2 2

2
2

2 1 2

2
__
c

_
=
g
a
_
2(1 +

2) (2 +

2)
(2 +

2) 1

2
__
c

_
= 0.
(152)
This gives us two equations relating c

and c

. However, these turn out to be the same equation,

so that only the relative relation between these constants is xed. From the rst row of eq. (152)
one has
c

=
2 +

2
2(1 +

2)
=
(2 +

2)(1

2)
2(1 +

2)(1

2)
=
1

2
, (153)
and thus
_
c

_
1

2
_
. (154)
This tells us that one normal frequency occurs if the upper pendulum swings to the left whilst
the lower one swings to the right (relative to the rst), with amplitudes of oscillation related as
shown. For the second solution, one substitutes
2
= g(2

2)/2a into eq. (149). Using similar

algebra to above, one nds that the other normal mode has the form
_
c

_
1

2
_
. (155)
Note this is similar to the other motion amplitude-wise. Now, however, the lower pendulum and
the upper pendulum both oscillate to the right. The normal frequency is lower for this mode
than for the other mode. It in fact turns out to be quite generally true that normal modes
which are more symmetric (i.e. here both pendula oscillating in the same direction) have lower
frequencies than modes which are less symmetric, or more antisymmetric (here exemplied by
the two pendula oscillating in opposite directions).
A general motion of the system is a superposition of the two normal modes. That is,
_

_
=
1
_
1

2
_
e
i1t
+
2
_
1

2
_
e
i2t
, (156)
where
1
and
2
are the two normal frequencies, and
1
and
2
constants which must be xed
from the boundary conditions. In practice, we take the real part of course
5
. Note that we can
5
The astute reader may notice that we had two second order dierential equations, and thus should have four
arbitrary constants in the general solution. Indeed I have ignored two constant phases which could also be present
in both exponents.
38
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
rewrite eq. (156) in the matrix form
_

_
=
_
1 1

2
__

1
e
i1t

2
e
i2t
_
, (157)
which may be inverted to give
_

1
e
i1t

2
e
i2t
_
=
_
1/2 1/2

2
1/2 1/2

2
__

_
=
_

1

2
_
, (158)
where we dened
j
=
j
e
ijt
. These are called normal coordinates. They are combinations of
the original coordinates ( and ) that have a single frequency associated with them. Looked at
another way, if we rewrite the Lagrange equations in terms of the normal coordinates, one nds
that the equations completely decouple, with a separate equation for each normal coordinate.
From eq. (158) one may write explicitly

1
=

2
+
1
2

2
,
2
=

2
+
1
2

2
. (159)
Then it is a reasonably straightforward exercise to show that the Lagrangian of eq. (141), written
in terms of the normal coordinates, is
L = ma
2
(2

2)

2
1
+ ma
2
(2 +

2)

2
2
2mga
2
1
2mga
2
2
, (160)
which indeed corresponds to two decoupled harmonic oscillators with dierent frequencies. The
Lagrange equations for
1
and
2
are

1
=
2
2

2
g
a

1
= (2 +

2)
g
a

1
(161)

2
=
2
2 +

2
g
a

2
= (2

2)
g
a

2
. (162)
These are harmonic oscillator equations with frequencies
2
= (2

2)g/a, as required. The

normal coordinates are perhaps the most natural coordinates to use in describing our system.
However, they do not necessarily have a clear physical interpretation at rst sight (like the
angles and do). This is one example of a point I made earlier in the course, namely that
generalised coordinates may be quite abstract in general, and not simply related to geometrical
properties of a diagram.
Above, we have seen an example of a physical system (the double pendulum) which when
displaced from its equilibrium position ( = = 0) undergoes small oscillations. It was found
that we can decompose any general oscillation of the system into two normal modes. Each of
these is a collective motion of the entire system with a single frequency, and there are as many
normal modes as there are generalised coordinates. In fact, this structure is quite general if we
perturb any dynamical system about a stable equilibrium. We take up the general theory of
small oscillations in the next section.
6.2 General theory of small oscillations
Consider a system with time-independent constraints. This is in equilibrium if
V
q
i
= 0 (163)
39
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
for all generalised coordinates q
i
. A stable equilibrium has

2
V
q
i
q
j
> 0 (164)
for all i and j. That is, if the potential is minimised then any forces which arise after displacing
the system from its equilibrium conguration act to restore the equilibrium, which is indeed
what we mean by stability. Denoting the equilibrium value of each generalised coordinate by
q

i
, we can expand each q
i
as
q
i
= q

i
+
i
, (165)
where
i
represents the displacement from the equilibrium value. Assuming that these displace-
ments are small, we can use Taylors theorem to expand the potential energy about q
i
= q

i
for
each coordinate:
V (q
1
, . . . q
n
) = V (q

1
, . . . q

n
) +
n

i=1

i
_
V
q
i
_

+
n

i=1
n

j=1
1
2
_

2
V
q
i
q
j
_

j
+ . . . , (166)
where we have neglected terms which are cubic (or higher) in the small displacements, and
the notation for the derivatives labels the fact that these are evaluated at the equilibrium
conguration. The rst term in eq. (166) is just a constant (i.e. it only depends on the constants
q

i
), and we can ignore this as it will not enter the equations of motion, which only depend upon
derivatives of V . Also, the second term is zero by the denition of the equilibrium conguration
(eq. (163)). Thus, for small displacements about equilbrium, we may take the potential as
having the form
V =
n

i=1
n

j=1
1
2
V
ij

j
, (167)
where we have dened
V
ij
=
_

2
V
q
i
q
j
_

. (168)
Note that this is an nn matrix (i.e. has two indices i and j). It is also symmetric (V
ij
= V
ji
),
which follows from its denition and the fact that it doesnt matter which order we take the
partial derivatives in. We will sometimes denote this matrix using the index-free notation V in
what follows i.e. V
ij
is the (i, j)
th
component of V. Then the set of displacements
i
forms a
vector , so that eq. (167) is equivalent to the matrix equation
V =
1
2

T
V, (169)
where
T
is the transpose of . As is clear from the denition of V, it does not depend on the
generalised coordinates (only on their equilibrium values), thus is a constant matrix.
Let us now consider the kinetic energy. In general, this will be quadratic in the velocities so
will have the form
T =
n

i=1
n

j=1
1
2
m
ij
q
i
q
j
, (170)
where, to be completely general, the coecients m
ij
may depend upon the generalised coordi-
nates. Again expanding the generalised coordinates about their equilibrium positions according
to eq. (165), one nds q
i
=
i
for each generalised velocity. Furthermore, one may expand the
coecients m
ij
as
m
ij
(q
1
, . . . q
n
) = m
ij
(q

1
, . . . q

n
) + . . . . (171)
40
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
Again we have used Taylors theorem. However, we only have to keep the zeroth order term
in this case, because the expression for the kinetic energy is already quadratic in the small
quantities
i
. Thus, the kinetic energy (for small displacements about equilibrium) can be
written
T =
n

i=1
n

j=1
1
2
T
ij

i

j
, (172)
where we have dened
T
ij
= m
ij
(q

1
, . . . q

n
). (173)
This is also a constant n n matrix. Furthermore, it can be chosen to be symmetric, as any
antisymmetric component will vanish after contracting with the symmetric combination
i

j
.
In matrix notation, eq. (173) reads
T =
1
2

T
T (174)
Having constructed the kinetic and potential energies, the Lagrangian for small displacements
about equilibrium is given by
L = T V =
n

i=1
n

j=1
1
2
[T
ij

i

j
V
ij

j
]

1
2

T
T
1
2

T
V. (175)
The above discussion is quite general, and tells us that any system, when displaced slightly
from equilibrium, can be represented by a Lagrangian having the form of eq. (175), in terms of
generalised coordinates
i
, which are related to the original generalised coordinates by eq. (165).
There are n Lagrange equations, one for each
i
. Each equation has the form
d
dt
_
L

k
_
=
L

k
. (176)
Considering the left-hand side rst, one has
L

k
=


k
_
_
1
2
n

i=1
n

j=1
T
ij

i

j
_
_
=
1
2

j
T
ij
_

i

k

j
+
i

j

k
_
,
where we have used the product rule in the second line. One may evaluate the derivatives using
the fact that the generalised velocities are independent, to give

i

k
=
ik
, (177)
where
ik
is the Kronecker symbol

ik
=
_
1, i = j
0, i = j
. (178)
Thus
L

k
=
1
2

j
T
ij
[
ik

j
+
i

jk
] .
41
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
The sum over i (j) in the rst (second) term can now be carried out, as the Kronecker symbol
picks out a single term in which i = k (j = k). One then gets
L

k
=
1
2

j
T
kj

j
+
1
2

i
T
ik

i
=

j
T
kj

j
,
where we have relabelled i j in the second term in the rst line, and also used the fact that
T
ij
= T
ji
. By a similar method, one may show that
L

k
=
n

j=1
V
kj

j
,
so that Lagranges equations for the system take the form
n

j=1
(T
ij

j
+ V
ij

j
) = 0. (179)
Or, in matrix notation
T +V = 0. (180)
This is an n-dimensional vector equation, as required.
As in the double pendulum case, we can look for solutions of the form

j
= c
j
e
it
, (181)
i.e. simple-harmonic-motion-like solutions involving some superposition of the coordinates os-
cillating with a single frequency. In vector notation this is
= ce
it
, (182)
where the vector c species how much of each coordinate we have in the oscillatory mode. Then
=
2
, and plugging this into eq. (180) and rearranging gives
_
V
2
T
_
= 0. (183)
For this equation to be satised, the determinant of the matrix which multiplies the vector
must vanish i.e.
|V
2
T| = 0. (184)
This is a polynomial of degree n for
2
, whose solutions are the normal frequencies (e.g. in
the double pendulum example we had two generalised coordinates ( and ), and there were
two solutions for
2
from the normal frequency equation). Eq. (184) is sometimes referred to
as a secular equation, and the procedure of nding normal frequencies is a bit like nding the
eigenvalues of a matrix. The only dierence is the fact that the matrix T is, in general, dierent
from the identity matrix.
Having found the normal frequencies, we can then substitute these back into the matrix
equation in turn, and nd the corresponding vectors c i.e. by solving
Vc =
2
Tc, (185)
42
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
for each
2
. This is a bit like nding the eigenvectors of a matrix, although not quite the same
because of the presence of T on the right-hand side. These vectors specify what we call the
normal modes of oscillation, as in the double pendulum example.
What can we say about the normal frequencies and modes in general? Firstly, it follows
from the fact that both V and T are symmetric that the solutions for
2
are real, and also that
the vectors c can be chosen to be orthogonal (hence the name normal modes). I will not
prove this here, but see e.g. the book by Goldstein et. al. if you are interested. There will be
n solutions for
2
, but some of these may be the same (i.e. there may be degenerate frequen-
cies). Also, the solutions for
2
are not necessarily positive - they may be zero or negative.
You will see an example of a zero frequency on the question sheet. We can understand what
negative frequencies correspond to as follows. A negative
2
results in a frequency which is
pure imaginary. Then, the complex exponentials in our trial solution actually turn out to be
real exponentials, so that the system runs away from equilibrium rather than oscillating around
it. Thus, negative solutions for
2
correspond to displacements about an unstable equilibrium.
It is to be expected that such solutions can occur, as we made no assumptions on the partial
derivatives of eq. (168) in the above analysis.
The general solution to the equations of motion takes the form
_
_
_

1
.
.
.

n
_
_
_ =

j
c
(j)
e
ijt
=

j
_
_
_
_
c
(j)
1
.
.
.
c
(j)
n
_
_
_
_
e
ijt
, (186)
where the sum is over all normal modes c
(j)
, each of which has normal frequency
j
. That is,
c
(j)
i
is the i
th
component of the j
th
normal mode vector. We can rewrite eq. (186) as a matrix
equation:
_
_
_

1
.
.
.

n
_
_
_ =
_
_
_
_
c
(1)
1
. . . c
(n)
1
.
.
.
.
.
.
.
.
.
c
(1)
n
. . . c
(n)
n
_
_
_
_
_
_
_

1
e
i1t
.
.
.

n
e
int
_
_
_, (187)
where the columns of the matrix consist of the various normal mode vectors. As in the double
pendulum case, we can then dene normal coordinates

j
=
j
e
ijt
, (188)
each of which corresponds to an oscillation of the system at a single frequency. These coordinates
can be found by inverting eq. (187) i.e.
_
_
_

1
.
.
.

n
_
_
_ =
_
_
_
_
c
(1)
1
. . . c
(n)
1
.
.
.
.
.
.
.
.
.
c
(1)
n
. . . c
(n)
n
_
_
_
_
1 _
_
_

1
.
.
.

n
_
_
_. (189)
In terms of these normal coordinates, the Lagrangian completely decouples into a sum of inde-
pendent harmonic osillators. That is,
L =

j
C
j
_

2
j

2
j

2
j
_
, (190)
where C
j
is a constant which depends upon the normalisation of the normal mode vector c
j
.
43
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m
That completes our discussion of the general theory of small oscillations. I have not given
many examples here, aside from the double pendulum example which I used to motivate the
general approach. There are a couple of examples on the question sheet, but it is worth briey
noting here some of the applications of normal modes:
1. Chemists are often interested in the vibrational energy spectra of molecules (i.e. there
are powerful spectroscopic techniques for identifying various compounds). Assuming the
bonds between atoms behave like springs, the vibrational energies are related to the normal
frequencies of a system of point masses joined by springs.
2. Musical instruments make noises by having something which vibrates to make sound waves
(e.g. the skin of a drum, the strings on a violin, air cavities in organ pipes). The pitches
of the notes one gets correspond directly to the normal frequencies of the vibrating object.
3. Many properties of elastic solids depend on the classical (and quantum) aspects of the
normal modes associated with vibrations of the crystal lattice. Applications abound in
condensed matter physics, but there are also more dramatic examples (e.g. earthquake
modelling!).
7 Closing remarks
In this course, we have seen a number of aspects of classical dynamics. We looked at rotating
frames and multiparticle systems in the Newtonian framework, before going on to introduce the
Lagrangian formalism. This allowed us to explore some of the most inspiring ideas in all of
physics (in my opinion at least), namely the principle of least action and the relationship be-
tween conservation laws and symmetries. Finally, we looked at an example of the application of
the Lagrangian approach, namely normal modes of oscillation. The quite general phenomenon
of small oscillations about stable equilibria occurs frequently in many dierent areas of physics
and chemistry.
I hope that I have managed to convey some of the power and underlying beauty of classical
dynamics. Needless to say, we have barely scratched the surface of this vast subject. Some of
the things we have not talked about are:
The motion of rigid bodies (continuous extended objects, rather than systems of point
particles).
The Hamiltonian approach to classical dynamics - it is possible to replace Lagranges
equations with a more symmetric looking set of equations involving the Hamiltonian H
of eq. (136). This shows up some very interesting mathematical structures underlying the
theory.
The relationship between classical and quantum mechanics. There are a number of fasci-
nating correspondences between classical and quantum dynamics, all of which make one
marvel at both theories!
The Lagrangian approach applied to relativistic systems.
The Lagrangian approach applied to elds rather than particles. This leads ultimately
to quantum eld theories, which are the current paradigm for describing nature at the
fundamental level. Strikingly, all fundamental particles are themselves seen to come from
elds.
Chaos theory!
I could go on of course, but the interested student will nd many excellent books on the above
subjects, as well as other lecture courses at this university...
44
w
w
w
.
k
i
n
i
n
d
i
a
.
c
o
m